Validation Method to Determine Metals in Atmospheric Particulate Matter by Inductively Coupled Plasma Optical Emission Spectrometry

Material particulado atmosférico (PM) é um poluente composto por diversos metais, que quando acumulados no sistema respiratório, podem causar sérios problemas de saúde. Os métodos IO-3.1 (extração de metais em PM) e IO-3.4 [determinação de metais por espectrometria de emissão atômica por plasma acoplado indutivamente (ICP-OES)] são os recomendados pela Agência de Proteção Ambiental dos Estados Unidos. Com o objetivo de avaliar o desempenho do método desenvolvido em nosso laboratório para extração de metais em PM utilizando HNO3 p.a. bidestilado e determinação por ICP-OES de Al, Ca, Cd, Cu, Cr, Fe, K, Mg, Mn, Na, Ni, Pb, Ti, V e Zn, a validação foi executada de acordo com os critérios estabelecidos pelo INMETRO, determinando parâmetros: seletividade, linearidade, precisão, exatidão, robustez, limites de detecção e quantificação, assim como a comparação da eficiência e precisão com o método IO-3.1. Resultados mostraram que o nosso método atendeu todos critérios de validação estabelecidos pelo INMETRO. Além disso, mostrou-se ser equivalente ao método IO-3.1.


Introduction
Atmospheric particulate matter (PM) pollutant is regulated in Brazil by the National Environment Consul through Resolution No 03/1990. 1 This resolution establishes the limit of daily exposure (average of 24 h) and continued exposure (annual average) to particulate matter in suspension (PTS) and inhalable particulate matter (PM 10 ).However, no limits were established for fine fraction (PM 2.5 ), therefore, studies are indispensable to create an inventory.
Since the eighties, researchers [2][3][4][5] had already noticed the great importance of studying toxic atmospheric pollutants, specially the determination of metals present in the particulate matter due to its direct relation with human health and the environment.Even in small concentrations, elements present in PM such as Zn, Cd, Cr, Co, Al, Fe, Mn, Ni, Pb and Cu present toxicological effects when bonded to short carbon atom chains or when they are present in their cation form and may form complexes that have a negative influence on the biological functions, affecting the normal operation of the tissues in the human body.These metals have non degradable characteristics; therefore accumulate easily in our organisms and ecosystems.
Besides the relation of metals present in particulate matter with health issues, it is possible to determine their connection with their source of origin, for example: natural, with the re-suspension of soil and marine aerosols and anthropogenic.Moreover, when receptor models are used, it is possible to identify if they are originated from fossil fuels operated by thermoelectric plants, steel mills, 2 or vehicles 6 which contribute to the degradation of air quality in a certain region. 7,8he determination of metal concentration in atmospheric particulate matter allows governments (federal, state and local) to execute control actions to reduce the impacts caused on human health.The advent of inductively coupled plasma (ICP) spectroscopy improved the performance and velocity of metal analysis in various ICP applications, being able to quantify metals on levels that are required by environmental laws. 9razil still has not defined the limit for metal concentration present in the air.Nevertheless, some countries already monitor and regulate metals present in particulate matter.Nonetheless, numerous methodologies are found in literature for the characterization of metals in PM. [6][7][8][9][10] The most used methodologies are the recommended by the United State Environmental Protect Agency (U.S. EPA): method IO-3.1 11 (extraction of metals in PM) and IO-3.4 9 [metal determination by inductively coupled plasma optical emission spectrometry (ICP-OES)].However, utilization of different analytical methodologies requires validation to ensure the efficiency of the method.
More and more laboratories seek to obtain analytical results that can be compared traceable and trusted otherwise they may lead to disastrous decisions and irreparable financial damage. 12Validation aims to demonstrate that an analytical method, in the conditions which it regularly takes place, possess the necessary characteristics for the obtainment of reliable and consistent results and contains the quality required for the application for which it is intended. 13,14According to National Health Surveillance Agency (ANVISA), 15 validation should ensure, through experimental studies, that the methodology meets the requirements of analytical applications, ensuring the reliability of the results.
Standardized methods, in general, have the following parameters pre-determined: selectivity, linearity, working range, precision, accuracy, recovery, robustness, limits of detection and quantification; validation execution is not required.However, analytical methods developed by the laboratory, methods not normalized, standard methods used outside of the scope for which it was designed or even extended or modified methods in relation to the standard do not have the parameters mentioned above.Therefore validation is required to ensure that the performance characteristics of the method meet the requirements for analytical operations intended, generating reliable and interpretable information about the sample. 12,13s well as the guides/resolutions available by national agencies for the validation of analytical methods, there are some defined/elaborated by international organizations such as International Conference on Harmonization (ICH) 16 for pharmaceutical appliances, the International Union of Pure and Applied Chemistry (IUPAC) 17 which registered a technical document that has been used by ISO 18 through the international norm ISO/IEC 17025 which is a specific norm for testing and calibration laboratories, and the United States Food and Drug Administration (US-FDA) 19 for foods and pharmacies in the United States.However, such agencies 13,15,16,18,19 and others require the validation of analytical methods item as a fundamental requirement for quality certification and demonstration of technical competence of the laboratory. 12he validation of a method is a continuous process that begins in the planning of the analytical strategy and continues throughout all its development.Thus, it is a long process and requires a large number of tests and statistical calculations resulting in a high cost analysis.Therefore, it is necessary to select the parameters that have a greater impact on the quality of the results and the speed in which they are obtained.1][22][23] The objective of this study was to validate a methodology developed in our lab in order to determine metals in particulate matter by ICP-OES.

Experimental
It is possible to find in literature various methodologies for the determination of metals in particulate matter. 6,7,9,10o evaluate the performance of the analytic procedure developed in our laboratory by Gioda et al. 24 to determine metals by ICP-OES, we selected some metals from different sources: Al, Fe, Ti, Mn, Mg (soil), Na, Ca (marine aerosol), Cd, Cu, Cr, Zn (industrial), K (biomass burning), Ni, Pb and V (vehicular).The metals were extracted from the PM with HNO 3 p.a. bidistilled and the validation was executed according to the criteria established by INMETRO. 13

Sampling
Fine particle matter (PM 2.5 ) samples were collected in downtown of the city of Rio de Janeiro (22º54'26.67''/43º11'43.13'')during 2011.The samplings were performed in 24 h periods, every six days, using high volume sampler and glass fiber filters, according to NBR 13412 method.An average flow of 1.16 m 3 min −1 was used.The PM 2.5 concentrations ranged from 12 to 40 µg m −3 , with an average of 25 ± 11 µg m −3 .The PM 2.5 mass collected in the filters ranged from 0.02 to 0.06 g.

Experimental procedure
The analytical procedure can be divided into two stages: acid extraction with bidistilled concentrated nitric acid and determination by ICP-OES of metals in atmospheric particulate matter.The careful execution of each step determines the quality of the results obtained in the process as a whole.

Metal extraction from atmospheric particulate matter
To accomplish the extraction of metals, 7 blank filters and 7 PM 2.5 samples were used.An aliquot (9 cm 2 , m ca.0.08 g) of each filter was cut with Teflon scissors, placed inside polypropene tubes and added 3.0 mL of HNO 3 (16.8mol L −1 ) bidistilled.These tubes were closed and heated for 2 h at 95 ± 5 °C.After that, the tubes were left to cool down to room temperature and were diluted to 25.0 mL with MilliQ water.Subsequently, they were centrifuged for approximately 5 min at a speed of 3,000 rpm so that the insoluble material was separated from the soluble.The soluble material was a supernatant solution of 12% v/v (2.0 mol L −1 ) HNO 3 which will be analyzed by ICP-OES.

Determination of metals in PM in the environment by ICP-OES
After the extractions, the metal content was determined by ICP-OES.The operating conditions (Table S1) and spectral lines (Table S2) are presented on Supplementary Information.
An external calibration curve of 0; 0.1; 0.2; 0.5 and 1.0 mg L −1 for Cd, Cr, Cu, Fe, Mn, Ni, Pb, Ti and V, and another curve of 0; 0.2; 0.5; 1.0; 2.0 e 5.0 mg L −1 for Al, Fe, Na, K, Mg, Zn and Ca were prepared.As the concentrations of Na, Ca, Mg, K and Zn are higher than the other metals mentioned, these were determined radially and the others axially.
To obtain the concentration of metals in particulate matter (mg L −1 ) it is necessary to multiply the results obtained by ICP-OES by the mass of the PM of the whole filter and divide it by the mass of the 9 cm 2 cut filter and subtract it from the results obtained for the blank filter multiplied by the mass of the whole filter and divided it by the mass of the cut blank filter.

Validation of the analytical method
The performance of the analytical method is influenced by the quality of instrumental measurements and the statistical reliability of the calculations involved in data processing. 25The selected parameters in the present study to test the performance of the methodology adopted, 24 according to the criteria established by INMETRO, 13 were: selectivity, linearity, precision, accuracy, robustness, limit of detection and limit of quantification, as well as the comparison of the efficiency of precision between methods of extraction.In addition, the tests were performed with certified reference material and primary standard which ensure the traceability of the results.
The sampling filter (glass fiber) has in its composition many analytes present in the atmospheric particulate matter.Then, for this study, the blank filter solutions obtained after acid extraction were considered as the sample matrix.Because its mass is ten times major than particulate matter.

Selectivity
The sample matrix contains, in its composition, many analytes present in the sample, which may interfere with the analytical signal of interest.The selectivity assesses the degree of interference of these components on the measured signal.The response obtained must correspond exclusively to the analyte. 26he selectivity can be obtained in various ways.A way to evaluate the selectivity, when it is not possible to obtain the matrix free of the substance of interest, is making use of the method of standard addition. 22Another way is to compare the matrix free from the substance of interest to the matrix added with this substance (standard) and evaluate the interference of the matrix with the matrix dissolution method. 12he first adopted methodology for the evaluation of selectivity was the standard to the matrix solution.For that matter, 8 blank filter samples were prepared, each one containing 9.0 mL of the standard solution and 7 of them containing the following volumes: 0.04; 0.08; 0.2; 0.4; 0.7; 0.8; 1.6 mL of the multielement standard solution of 50 mg L −1 and the concentration of each element was analyzed by ICP-OES.
To evaluate the selectivity using this method, residual graphs were plotted and the Pearson correlation coefficients were calculated according to equation 1. (1) where, x i and y i correspond to the values of the variables X and Y;x andy are the mean values of x i e y i .
The second method used to identify the possibility of interference from the matrix was the dissolution of the matrix.To this end, 7 PM 2.5 samples were prepared with 5.0 mL of matrix solution diluted in the following proportions: 1, 1:1, 1:2, 1:4, 1:10, 1:20, 1:40, and to other seven identical samples were added 0.5 mL of standard solution multielement 50 mg L −1 .All samples were analyzed by ICP-OES to determine the concentrations of each element studied.
In addition, the t-student test was carried out to make sure if the results with different dilutions of the matrix solution influenced the analytical response when standard solution was added.
To verify the hypothesis test, a level of trust of 95% and 4 degrees of freedom were considered to evaluate the difference in concentration with and without the addition of the standard to the matrix solutions with different dilutions.

Linearity
The simplest way to describe linear dependence 16 is by the linear regression model, represented by a mathematical equation (equation 2) where y is a function of x.
The adequacy for the adjustment of the curve 27 is provided by the Pearson correlation coefficient (r).

Limit of detection and quantification
The concentrations of each analyte from the matrix solution were determined from 10 readings, and the mean (µ) and sample standard deviation (s) were obtained.The limit of detection (LOD) was determined from these results by applying equation 3 and also the limit of quantification (LOQ) with the use of equation 4.

Recovery
The recovery or recovery factor (R) is defined as the ratio of the amount of substance of interest added or present in the sample, recovered by means of the analytical method. 17It therefore refers to an amount of specific analyte effectively quantified in relation to the "real" amount present in the sample.
To evaluate the recovery of our method 24 two identical samples of CRM were prepared.For each sample, acidic extraction of 10 mg of CRM of urban particulate matter (NIST SRM 1648a) 28 was prepared using 3.6 mL of concentrated bidistilled nitric acid (16.8 mol L −1 ), and heating at 95 ± 5 °C for 2 h.Therefore, the volume was completed to 30.0 mL with nanopure water, resulting in a solution with a final concentration of 12% v/v HNO 3 (2.0 mol L −1 ).The supernatant was analyzed by ICP-OES.The results observed for each analyte were evaluated by comparing them to certified values for SRM 1648a (expected value) and the efficiency of the determined method (equation 5).
The expression below refers to the calculation of the recovery factor (equation 5): Vol. 25, No. 9, 2014 (5)   To assess the recovery factor was used an average of the two results of the observed values.

Accuracy
To evaluate the accuracy 29 of the method the relative error (ER) was calculated according to the following expression (equation 6): (6)   where X lab is an average of the results obtained from the 3 parallel samples of standard solution and X v is the expected value.
The accuracy was obtained by the evaluation of 4 multielement standard solutions with concentrations belonging to the working range: 0.2, 0.5, 1.0 and 5.0 mg L −1 , added to the matrix solution.

Precision
The precision was evaluated by the relative standard deviation (RSD).For this work intra-run precision (repeatability) and inter-run precision (intermediate precision) were made.

Repeatability
The repeatability of the method was examined for a range of multielement standard solution concentrations of 0.2, 0.5, 1.0, 2.0 and 10.0 mg L −1 added to the matrix solution.The mean (m) and standard deviation (s) were calculated based on the results obtained for each concentration to calculate the RSD (%) (equation 7). (7)

Intermediate precision
The intermediate precision refers to the precision evaluated upon the same standard, using the same method, in the same laboratory, but defining one or more conditions.In this study, the days of analysis varied as well as the analyst.
The multielement standard solutions with 0.2, 0.5, 1.0, 2.0 and 10.0 mg L −1 added to the matrix solution were reanalyzed after two days and reassessed by a different analyst and the RSD was calculated (%) (equation 8).(8)   where s 1 and s 2 are the standard deviations, m 1 and m 2 are the averages of 7 replicas of the first and second day of analysis of standard solutions, respectively.

Comparison of precision and recovery from other extraction methods
For the evaluation of the precision and recovery of the extraction method used in our lab, 24 the method U.S. EPA IO-3.1 11 was adopted as the reference method.
All samples used for the evaluation of the acid extraction method in the lab 24 were the ones prepared for the evaluation of the recovery.However, the sample used for the evaluation of the extraction using the U.S. EPA IO-3.1 method was prepared with 10 mg of certified reference material 28 (NIST SRM 1648a) in 0.9 mL of HNO 3 (16.8mol L −1 ) and 2.4 mL of HCl (12.0 mol L −1 ), with heating for 30 min, and the volume completed to 30 mL, resulting in a supernatant solution of concentration of 3% HNO 3 /8% HCl (1.46 mol L −1 ).
The calculation for the recovery factor was done by considering the observed value as the average of the results of the two parallel samples.Once the recoveries from both extraction methods mentioned above were known, an analysis of their differences was made and their equivalence evaluated.Besides the evaluation of the efficiency of extraction, the precision of both methods was determined to see whether they presented significant differences with the application of the F-test for a confidence level of 95%.

Robustness
With the purpose to verify if the method is robust to small deliberate variations of the optimized conditions, the robust test proposed by Youden and Steiner 30 was done in a sample of particulate matter, which was divided into eight equal parts, collected in the city of Rio de Janeiro.
The test consists of a multivariate analysis of seven variables that may influence in the analytic result using the results observed in eight experiments, varying the extraction and determination conditions.The seven variables selected for the execution of the Youden and Steiner test 30 are shown in Table S3 (Supplementary Information), as well as the normal experimental procedure used, which are represented by upper case letters A, B, C, D, E, F, G and representing the alternative conditions the corresponding lower case letters a, b, c, d, e, f, g.
From these results, the effect of each variable is estimated by the difference between the average results of four analyses with an uppercase letter and the average results of four analyses with lowercase letter.Absolut effect values greater than the standard deviation multiplied by the square root of two (s√2) were considered significant and thus alter the analytical response.

Selectivity
The first evaluation of selectivity was performed by a standard addition curve.This was prepared by adding acid to blank filter (matrix solution) with volumes of 0.04 to 1.6 mL of multielement standard solution 50 mg L −1 foreseeing concentrations 0.2 to 8.0 mg L −1 , which are concentrations that are commonly found in atmospheric particulate matter.
The standard addition curve was divided into two parts; the first (1) contained the standard in low concentrations (0 to 1.0 mg L −1 ) and the other (2) in medium concentrations (1.0 to 8.0 mg L −1 ).The second order polynomial regression was applied on the results of two standard addition curves and calculated correlation coefficient (Pearson), as shown in Table S2 (Supplementary Information).
The curves appear to be quite linear, with the Pearson coefficient ranging from 0.9978 to 0.9999, satisfying the recommended by ANVISA 15 (r ≥ 0.99) as well as the required by INMETRO 13 (r ≥ 0.90), as evidence of an optimal fit of the data to the regression line.
Therefore, all the correlations can be considered very strong, ensuring a good correlation.To analyze the consistency of the standard addition curves an assessment of the residue data from the 2 nd order polynomial regression was done in relation to the experimental values (Figure 1).
Figure 1 indicates that the residues were mostly less than ± 2%, only a few exceeded this value, however, they did not exceed 10%, indicating that the 2 nd order polynomial regression is a good representation of the concentrations found in the standard addition curve.Therefore, it can be concluded that although the sample filter of the atmospheric particulate matter (sample matrix) contains in its composition many of analytes present in the sample, they do not interfere with the analytical signal of interest.The second evaluation of selectivity was done to assess if there was any interference of the matrix in the analytic results of the particulate matter.A t-student test (Table S4, Supplementary Information) was done using results of the differences into metal concentrations of dilutions of the matrix with and absent addition of standard solution.
As the concentrations results obtained for several dilutions revealed to be equivalent for all investigated elements, it can be concluded that the dilution of the matrix does not influence the analytical response.Therefore, it is not necessary to dilute the sample of particulate matter to dilute the matrix.Thus, the manner in which the sample was used for the determination of metals by Gioda et al. 24 was sufficiently adequate.A procedure to select which optimum dilution of the matrix should be adopted for the determination of metals in particulate matter by ICP-OES is by observing the limits of detection of each dilution as shown in Table 1.
The lower limits of detection for each of the analyzed metals with the different dilutions of the matrix are highlighted in gray in Table 1.
It can be concluded that the way in which the acidic solution from the extraction of particulate matter by Gioda et al. 24 for the determination of metals by ICP-OES (without dilution of the matrix) was the best procedure that could have been adopted as, for most metals when dilution was not made, a lower limit of detection was obtained.

Linearity
To evaluate the linearity, three external calibration curves were prepared: 0.01 to 0.1 (1), 0.2 to 1.0 (2), 2.0 to 10.0 (3) mg L -1 of multielement standard solution.The heteroscedasticity was evaluated through the residue graph (Figure 2) in relation to the linear regression.Extremely high residues were observed with an average of 133% when a single curve was used for this range of concentration.However, when the residues were checked using a curve for low concentrations (0.01-0.1 mg L −1 ), another curve for medium concentrations (0.2-1.0 mg L −1 ) and one for high concentrations (2.0-10.0mg L −1 ) the residues were to ± 2% (Figure 2).Only calcium and lead obtained a residue greater to 10%.
Table S5 (Supplementary Information) shows the Pearson coefficients of low, medium and high concentration curves when done the linear regressions following the of least squares method of the analyte concentration versus the emission intensity.
According to Green 32 and Shabir, 33 correlation coefficients values higher than 0.999 is the evidence of an optimal fit of the data for the regression line.However, the curves were quite linear for all elements in the concentrations evaluated, with the exception of the elements K and Na, which only showed a correlation coefficient upper to 0.999 after the concentration of 1.0 and 0.1 mg L −1 , respectively.All correlation coefficient values of K and Na were higher than that recommended by ANVISA 15 (r ≥ 0.99) and the required by INMETRO 13 (r ≥ 0.90), considering that the curve is linear.

Limit of detection and quantification
The limits of detection and quantification shown in Table 2 were determined with the matrix solution.The limits of detection were inferior to 0.02 mg L −1 for most of the metals with the exception of Al, Ca, Fe, K, Mg, Na e Zn that remained in the range of 0.1 a 16.0 mg L −1 as they are already present in significant concentrations in the blank filter.Regarding the detection limits described in U.S. EPA Method IO-3.4 for metals that are not heavily present in the matrix of the sample, very close values can be observed (Table 3).

Recovery
To evaluate the recovery (Table 4) of the method of extraction, solutions with SRM1648a were prepared with 10 mg, to ensure the homogeneity of the sample and to approach the mass of the sampled atmospheric particulate matter.
Once the recovery of the method is known, it can be used and the recovery factor can be applied to the results to know the total concentration of all analytes present in the samples of particulate matter.

Accuracy
The accuracy was obtained by the evaluation of standard solution for four concentrations belonging to the working range: 0.2, 0.5, 1.0 and 5.0 mg L −1 , added to the matrix solution.The relative error was calculated in relation to the experimental results and the expected results (Table 5).
The relative errors varied from 0.2 to 10% for all elements.Therefore, they are below the maximum recommended by ANVISA 15 (15%).

Repeatability
The repeatability of the method was examined for a range of multi-element standard solution of concentrations 0.2,  0.5, 1.0, 2.0 and 10.0 mg L −1 that were added to the matrix solution.The results of the relative standard deviation for each of the 15 elements determined are shown in Table 6.

Intermediate precision
Intermediate precision was evaluated on the same sample used to evaluate the repeatability, using the same method in the same laboratory, but in different days and with different analysts.Table S6 (Supplementary Information) presents the RSD of intermediate precision at different concentrations of analytes.
According to Huber, 21 the methods used to quantify the analyte with maximum of 10 mg L -1 has an acceptable RSD lower or equal to 7.3% and for analytes of 1 mg L −1 lower or equal to 11%, depending on the complexity of the sample.
On the other hand, ANVISA 15 establishes the maximum as 15%.Therefore, the relative standard deviations observed in the analysis of repeatability and intermediate precision are satisfactory for the concentration range examined, as they were always inferior to 5%.

Comparison of the precisions and recoveries of extraction methods
After determining the concentrations of the analytes of two extraction methods, U.S. EPA IO-3.1 11 and our method, with duplicated samples (n = 2), for the same mass of certified reference material SRM 1648A (m = 10 mg), by the same method of determination (ICP-OES) under the same analytical conditions, the analyses of the differences of efficiencies of extraction (recovery) (Table 7) and an F-test (precision) were done to assess if there is any equivalence between the methods (Table 14).
The recoveries are very similar, not differing by 5% for most metals.This value was only exceeded for Cd, Fe and Mn, it however, did not exceed 15%.
It is important to remark that as our as U.S. EPA IO-3.1 methodologies only determine the metals soluble in HNO 3 , excluding the fraction of metals binding to silicates.
The results of the F-test indicated that the two extraction methods have equivalent precisions (Table 8).Therefore, the method adopted in our lab is quite similar to U.S. EPA method IO-3.1, which is the most common method for the determination of metals in atmospheric particulate material.

Robustness
Comparing the results of the effects of each variable to the standard deviation of 8 samples multiplied by the square root of two (√2 s) it was impossible to evaluate whether the values of the effects were significant and therefore alter the analytical response (Table 9) .
Applying the Youden and Steiner test, 30 it can be concluded that only Ni may suffer variations in its analytical response if the power of the equipment varies from 500 W to 400 W, which is not expected during the analysis due to the stability of the plasma in this equipment.Therefore, we can conclude that the method is robust for all variables found in the considered experimental domain.

Conclusion
Among the analyzed elements, Al, Na, K, Ca and Zn are strongly present in the matrix of the sample.However, through tests of selectivity, it can be observed

Figure 1 .
Figure 1.Residue of (a) Al, K, Mn and Na; (b) Ca, Zn, Ni and Pb; (c) Cd, Cu, Cr and Fe and (d) Mg, Ti and V obtained of the 2 nd order polynomial regression of the standard addition curve.

Figure 2 .
Figure 2. Residues obtained of (a) Al, Cd, Cu and Cr; (b) Fe, Mg, Mn and Ni; (c) Ca, K, Na and Pb and (d) Ti, V and Zn by using 3 external calibration curves.

Table 1 .
Limit of detection (mg L −1 ) in several dilutions of the matrix

Table 2 .
Limits of detection and operational quantification (mg L −1 ) for

Table 4 .
Analytical recovery with SRM 1648a

Table 5 .
Relative errors (%) per range of analyte concentration

Table 6 .
Relative standard deviation (%) in different analyte concentrations

Table 8 .
Equivalence test between the precision of the extraction methods